Generative Adversarial Nets for Multiple Text Corpora

TL;DR

This paper explores the use of GANs to generate cross-corpus word and document embeddings from multiple text corpora, improving supervised learning tasks.

Contribution

It introduces novel GAN models tailored for multiple text corpora, enabling consistent cross-corpus embeddings and robust document representations.

Findings

GANs improve cross-corpus word embedding consistency

GAN-generated document embeddings enhance supervised learning

Models demonstrate effectiveness on real-world datasets

Abstract

Generative adversarial nets (GANs) have been successfully applied to the artificial generation of image data. In terms of text data, much has been done on the artificial generation of natural language from a single corpus. We consider multiple text corpora as the input data, for which there can be two applications of GANs: (1) the creation of consistent cross-corpus word embeddings given different word embeddings per corpus; (2) the generation of robust bag-of-words document embeddings for each corpora. We demonstrate our GAN models on real-world text data sets from different corpora, and show that embeddings from both models lead to improvements in supervised learning problems.